Gemini Omni 1.2 Flash: benchmarks, pricing and the business video use case

The expensive part of business AI video is rarely the first generation. It is the repeated attempts that alter the product, distort packaging or replace the scene whenever you request a correction. Gemini Omni 1.2 Flash deserves attention for combining quality, reference adherence and editing of an existing clip.
Google announced it on November 5, 2026: clips up to 15 seconds, conversational editing and improved reference fidelity. Its quoted 720p output price is approximately $0.07 per second, 30% below Omni 1.1. Google announcement
The useful business question is how much an approved clip costs.
Benchmarks: a strong, preliminary start
Arena’s November 6 snapshot places Omni 1.2 first in both text-to-video and image-to-video.
| Model | Text → video | Image → video |
|---|---|---|
| Gemini Omni 1.2 Flash | 1541 ± 15 | 1502 ± 12 |
| Gemini Omni 1.1 Flash | 1514 ± 15 | 1486 ± 11 |
| Dreamina Seedance 2.5 (720p) | 1480 ± 8 | 1488 ± 7 |
| Wan 3.0 | 1477 ± 12 | 1481 ± 11 |
| Grok Imagine Video 1.5 | 1486 ± 18 | 1454 ± 4 |
| Veo 3.1 (Audio) | 1363 ± 14 | 1399 ± 10 |
Sources: Text-to-Video Arena and Image-to-Video Arena. Scores across columns are not directly comparable. These are separate evaluations and the model variants follow the leaderboard labels; Grok 1.5 is labelled 720p in image-to-video.
Both Omni results are Preliminary, based on 1,742 text-to-video votes and 2,745 image-to-video votes. MiniMax H3 scores 1493 ± 5 in image-to-video with over 121,000 votes. Leading intervals overlap; Arena gives Omni an image-to-video rank spread of 1–6.
This earns Omni a place on the shortlist, not an unconditional recommendation. Arena measures preferences between outputs, not packaging accuracy, Czech speech quality, correction count or editor time. This article compares public evidence; it is not our own hands-on benchmark.
Where it stands out
Product and character consistency. Google reports stronger adherence to image and video references. For an online store, the potential benefit is fewer shots that change a garment or product when the camera moves. This is a vendor claim, not a guarantee. I would test one product in three settings and score fidelity to the original.
Corrections without starting from scratch. Conversational editing can revise lighting or camera angle in an existing clip. It saves work only when the correction preserves what already worked. Another edit is not automatically free.
Separate draft and finishing budgets. Google recommends drafts at 360p or 720p, then upscaling selected results to 1080p or 4K. Upscaling is not proof of native 4K detail. It is nevertheless a sensible way to separate exploration from finishing. Google Cloud
Pricing: compare the same resolution
Public rates checked November 7, 2026, for 720p output in USD. Input media, edits, upscaling, taxes and human work are excluded.
| Model | USD / second at 720p | 100 seconds of output |
|---|---|---|
| Grok Imagine Video 1.5 Lite | 0.03 | $3 |
| Gemini Omni 1.2 Flash | approx. 0.07 | approx. $7 |
| Grok Imagine Video | 0.07 | $7 |
| Grok Imagine Video 1.5 | 0.14 | $14 |
Sources: Google Cloud and xAI pricing. The 100 seconds are accumulated output, not a single-clip duration claim. This is a selected price comparison, not pricing for every benchmark entry.
Grok Imagine Video 1.5’s $0.08/second rate is 480p. At 720p, Omni’s quoted output rate is roughly half the full Grok 1.5 rate. Lite is a different model and cannot inherit the full model’s benchmark scores.
OpenRouter’s starting Omni price of roughly $0.02317/second is likewise 360p. Its $1.50/$9 per million input/output token headline is not a complete video price; video has separate resolution-dependent rates. OpenRouter
Omni is not the cheapest option in every configuration. Its combination of pricing and current results is the attraction.
The cost of an approved clip
Using the rounded $0.07/second rate, a 15-second 720p output costs approximately $1.05. One hundred approved clips would cost $105 with one attempt each, $315 with three attempts, or $525 with five. These are output-only arithmetic examples, not measured model acceptance rates.
Cost per approved clip = all attempt and processing costs / approved clips.
A cheaper model with a one-in-five acceptance rate may cost more in practice. A better Arena score does not compensate for changing your product.
A practical production pilot
Start with twenty products, good reference photos and approved descriptions. An n8n or Make workflow can retrieve assets, build a prompt from a template, call the video API and store the output alongside its generation settings. A human approves drafts before selected clips proceed to finishing.
Track product fidelity (shape, colour, packaging and text), motion quality, generation and edit counts, and time to approval including review. Set a per-product retry and spending limit to prevent endless unsuccessful generation.
Run identical briefs through Grok Imagine Video 1.5 and another available Arena contender. Without shared inputs and scoring rules, it is a showcase rather than a benchmark.
For onboarding and process explanations, I would still favour real screen recordings, a verified script and Google Vids. Generated footage fits supporting scenes, product variations and short advertising formats better.
Verdict
Omni 1.2 Flash is a strong candidate for repeatable short product-video production. Its potential is reaching an approved clip from reference photos with fewer wasted attempts. Test that claim in your own pilot. Arena gives you a reason to evaluate it; your accepted outputs should decide deployment.
Sources and timing: checked November 7, 2026; Arena snapshot November 6. Google: release, Google Cloud, Arena text-to-video, Arena image-to-video, xAI pricing, OpenRouter. Editorial signal: AI Revolution CZ/SK, November 7 video.